My Portfolio#

CANIO#

Developed NJIT Solar Car’s CAN IO board, to interface the car’s various peripherals with the car’s CAN bus. Uses an STM32G0B1 microcontroller, as well as isolated CAN Transceiver to interact with the 1Mbps CAN 2.0A bus.

Currently writing firmware for both the CAN IO board, as well as the car’s STM32F413-based central control unit. Find the car at the FSGP competition this July in Minnesota!

Go Graphics Virtual Machine#

Developing a stack based virtual machine that supports drawing images, by wrapping the GG library’s functions. Effectively, draw images using the GG library without needing to recompile Go code. Lightweight (currently 3MB compiled), somewhat fast, and very nearly turing complete (only need indexable variables, coming soon!).

• Currently, the core VM is complete as initially intended. As development progressed, I saw a few extra features it could benefit from (ie, indexable variables; allowing stack reversing functions, and turing completeness).
  • Currently supports stack pushing and popping, loading and storing from a hashmap-based memory (store values from stack to memory based on variable name, not memory address), branching, calling functions (with return address), and importing other files, with naive import loop checking.
  • Currently supports almost half of the GG library’s functions; I plan on finishing just the bare, orthogonal instructions as soon as possible, with the full instruction set possibly at some point in the future.

ESP32 Bootloader#

Building a toy bootloader for an ESP32-S3. Inspired by the toy Rust bootloader created by The Rusty Bits on Youtube. Still in very early design phase, but I plan on implementing support for at least a few peripherals (UART, maybe I2C), and also possibly an over-the-air firmware update system via an RF module.

Autonomous Lawnmower#

Group Engineering Capstone project. An autonomous navigation system for lawnmower, utilizing YOLOv8 segmentation, LiDAR, and Ultra Wideband Positioning (UWB) for real-time localization and path planning.

The project boasts completely custom hardware and firmware:

STM32F446-based Logic Analyzer#

Group project for a Masters Embedded Systems course. STM32F446 based Logic analyzer. Capable of sniffing upto 500kHz signals.

• DMA data acquisition, and run-length encoded data transmission, to optimize byte usage on USB 2.0 FS
• UART, SPI, I2C, and CAN signal decoding via a custom Go backend.
• GUI using PyQt5. Sends and receives data from the STM via the Go backend, using a socket connection.

BitSum Lecture Summarizer#

A completely local, docker based AI stack which takes input of audio/video lectures, and transcribes and summarizes them. Uses OpenAI Whisper for transcription, and Microsoft BitNet for efficient CPU-based summaries.

NeuralNet#

A C library to create, train, and export feed-forward neural networks. It leverages a completely custom (non-hardware accelerated) matrix library for matrix/vector operations, as well as MSE Loss, SGD based backpropagation (using matrix derivatives), and ReLU activation.

SessionSummary Email Generator#

Created a fully custom email generator for the Mathnasium of Chatham + 4 other centers. Utilizes Javascript and Node.js, and custom (effectively reverse-BNF grammar) email generation algorithm to generate 110-150 emails per day.